[go: up one dir, main page]

WO2025064557A1 - Reconnaissance précoce de changement d'état pathophysiologique de dysglycémie - Google Patents

Reconnaissance précoce de changement d'état pathophysiologique de dysglycémie Download PDF

Info

Publication number
WO2025064557A1
WO2025064557A1 PCT/US2024/047305 US2024047305W WO2025064557A1 WO 2025064557 A1 WO2025064557 A1 WO 2025064557A1 US 2024047305 W US2024047305 W US 2024047305W WO 2025064557 A1 WO2025064557 A1 WO 2025064557A1
Authority
WO
WIPO (PCT)
Prior art keywords
data
model
person
dysglycemia
glucose monitoring
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
PCT/US2024/047305
Other languages
English (en)
Inventor
Richard A. Frank
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Frank Healthcare Advisors LLC
Original Assignee
Frank Healthcare Advisors LLC
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Frank Healthcare Advisors LLC filed Critical Frank Healthcare Advisors LLC
Publication of WO2025064557A1 publication Critical patent/WO2025064557A1/fr
Pending legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H10/00—ICT specially adapted for the handling or processing of patient-related medical or healthcare data
    • G16H10/60—ICT specially adapted for the handling or processing of patient-related medical or healthcare data for patient-specific data, e.g. for electronic patient records
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/145—Measuring characteristics of blood in vivo, e.g. gas concentration or pH-value ; Measuring characteristics of body fluids or tissues, e.g. interstitial fluid or cerebral tissue
    • A61B5/14532—Measuring characteristics of blood in vivo, e.g. gas concentration or pH-value ; Measuring characteristics of body fluids or tissues, e.g. interstitial fluid or cerebral tissue for measuring glucose, e.g. by tissue impedance measurement
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7271—Specific aspects of physiological measurement analysis
    • A61B5/7275—Determining trends in physiological measurement data; Predicting development of a medical condition based on physiological measurements, e.g. determining a risk factor
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H20/00—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance
    • G16H20/10—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance relating to drugs or medications, e.g. for ensuring correct administration to patients
    • G16H20/17—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance relating to drugs or medications, e.g. for ensuring correct administration to patients delivered via infusion or injection
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H40/00—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices
    • G16H40/60—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices for the operation of medical equipment or devices
    • G16H40/63—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices for the operation of medical equipment or devices for local operation
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H40/00—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices
    • G16H40/60—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices for the operation of medical equipment or devices
    • G16H40/67—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices for the operation of medical equipment or devices for remote operation
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/30—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for calculating health indices; for individual health risk assessment
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/50—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for simulation or modelling of medical disorders

Definitions

  • Continuous glucose monitoring data reveals patterns of glycemia, which vary from one person to another and are affected by meals and snacks, exercise and sleep, and other factors. These patterns can be identified for numerous days to train a person- specific model.
  • the model can take in a real-time pattern or time of day information or additional related information such as meals, exercise, and sleep.
  • Identifying the magnitude and changes over time of deviations between models of glycemia patterns within a person over certain time periods indicates stability or failure to maintain the pattern of glycemia predicted by the initial models, indicating the onset of, or worsening of, pathophysiologic dysglycemia.
  • FIG. 1A shows a chart of a scenario of an individual person's glucose concentration variation throughout a day with three meals.
  • FIG. 1 B shows a chart of a scenario of a multiplicity of glucose concentration variations throughout multiple days for an individual person.
  • FIG. 2 shows an example of a continuous glucose monitor that fits on a person's wrist.
  • FIG. 3 shows an example of a continuous glucose monitor that sticks to the skin and can be worn under clothing.
  • FIG. 4 is a diagram of examples of connected devices connectivity.
  • FIG. 5 is a diagram of an example of a method for reporting pathophysiologic dysglycemia.
  • FIG. 6 is a diagram of an example of a method for reporting pathophysiologic dysglycemia.
  • FIG. 7 is a table of types of values for examples of risk factors on which a predictive model can be conditional.
  • FIG. 8 is a diagram of an example of a method for reporting pathophysiologic dysglycemia.
  • FIG. 9 is a diagram of an example of a method for reporting pathophysiologic dysglycemia.
  • FIG. 10 shows examples of standard of care pathways.
  • FIG. 11 shows an example of a standard of care pathway using detection of deviation, interpretation, and a conclusion of diagnosis and recommended intervention.
  • DKA diabetic ketoacidosis
  • Type 1 diabetes is a chronic metabolic disorder characterized by insufficient insulin production by the pancreas. Insulin promotes the absorption of glucose from the blood into cells of the body. A lack of insulin causes the glucose homeostasis to change from a normal state (euglycemia) to an abnormal state (dysglycemia).
  • euglycemia a normal state
  • dysglycemia a lack of insulin causes the glucose homeostasis to change from a normal state (euglycemia) to an abnormal state (dysglycemia).
  • Hypoglycemia is commonly defined as a blood sugar concentration lower than 70 milligrams per deciliter (mg/dL), whereas hyperglycemia is commonly defined as a blood sugar concentration higher than 200 mg/dL.
  • Type 1 diabetes can lead to ketoacidosis, heart disease, neuropathy, nephropathy, retinopathy, foot damage, skin infections, pregnancy complications, among other complications.
  • Current treatment methods involve lifestyle modifications, insulin therapy, medications, artificial pancreas (a closed-loop insulin pump), beta cell replacement therapies, and others. None of these approaches are fully effective, and most have serious side effects. There is no known cure for type 1 diabetes.
  • dysglycemia is a criterion for Stage 2, and is also one of the criteria used by physicians in prescribing the preventative therapy which can delay onset.
  • OGTT Oral Glucose Tolerance Test
  • OGTT requires a clinic visit and at least two blood draws hours apart. That process is inconvenient and time-consuming for patients and creates a slower and therefore more expensive workflow for clinics. It is administered at long intervals and misses the earliest opportunity for intervention to preserve insulin secretion.
  • AICGM can determine which children should be tested using the OGTT.
  • AICGM can enable a change in the standard of care by replacing OGTT, which is cumbersome and episodic. Patients, especially in the general population, often skip tests. Under any circumstances, OGTT only detects the instantaneous state of glycemia. If the interval of testing is not coincident with the early periods of dysglycemia (“interval incidence”), the patient may suffer complications of diabetes shortly after the last. OGTT and months or years before the next one is scheduled. Methods disclosed herein can enable early detection, diagnosis, and intervention to delay the onset of insulin dependence.
  • AICGM in some cases, can supplant the conventional step of referring patients to an OGTT test. In such a case, a patient can be referred directly to a course of treatment. Such a use could be manifest in Clinical Practice Guidelines and United States Preventive Services Task Force (USPSTF) screening recommendations.
  • USPSTF United States Preventive Services Task Force
  • the Juvenile Diabetes Research Foundation advocates screening the general pediatric population for the risk of diabetes.
  • AICGM can be instrumental in optimizing such screening by providing the earliest detection of dysglycemia warranting intervention to delay die onset of insulin dependence.
  • Type 1 diabetes can be caused by an autoimmune attack on pancreatic beta cells. Beta cells produce insulin for regulating blood glycemia. Beta cells also produce glutamic acid decarboxylase (GAD). Type 1 diabetes caused by an autoimmune attack that kills beta cells occurs when the immune cells respond to GAD as an antigen. A blood test that detects immune cell response to GAD indicates that the person is, so called, antibody positive (Aby+). Otherwise, the person is antibody negative (Aby-). There are 3 other antibodies which can contribute to tire autoimmune atack on insulin- producing beta cells. The detection of dysglycemia, as described herein, in a person with any two of these four antibodies, is diagnostic of Stage 2 diabetes and warrants intervention with the preventative therapy.
  • GAD glutamic acid decarboxylase
  • a person who is known to be Aby-f- has a different prior risk of having type 1 diabetes than someone who is Aby-.
  • a person with a 1st degree relative who is diabetic also has a different risk than someone in the general population. There also are other genetic risk factors.
  • Preventative intervention is distinct from a therapeutic.
  • the drug Tzield is like a vaccine, preventing the immunologic destruction of insulin -producing beta cells, thereby delaying insulin dependence.
  • An older approach is therapeutic.
  • it is prescription of exogenous insulin in compensation for the patient’s pathophysiologic insufficiency which results from prolonged autoimrnune-mediated destruction of insulin-producing beta cells and, hence, insulin dependence.
  • FIG. IB shows a chart of glucose measurements across a full day for several days. Whether they have pathologic dysglycemia or not, to what degree, and whether the condition is deteriorating is essentially impossible to discern from the data by direct human observation.
  • a model of an expected pattern of glycemia can be created by analysis of a large number of patterns using machine learning. Such a model can be implemented, for example, as a neural network or other model having parameters that give specifically calculated weights to input data. Some such neural networks use recurrence. Simple recurrence is useful for representing instantaneous changes in glycemia from one time sample to the next.
  • a more complex form of recurrence is a long short-term memory (LSTM). This can be useful to represent the fact that glucose is, in essence, stored in the blood.
  • the size of the body and organs can affect the storage. So, the rate of change of glycemia can have latencies based on the amount of stored glucose.
  • the storage can be modeled by a LSTM.
  • One or more models can be trained to predict one or more of the likelihood of type 1 diabetes, severity of episodes of dysglycemia, maximum glycemia, and responsiveness to acute or ongoing treatments. Methods of training such models will be described below.
  • Glycemia can be determined usings assays or electronic sensors. Glycemia can be measured directly or estimated by secondary indications through analysis of blood, intracellular fluids, interstitial fluids (ISF), tears, saliva, urine, sweat, and other fluids inside or exuded from the body.
  • ISF interstitial fluids
  • Various types of glucose monitoring devices can be assistive, augmentative, or autonomous.
  • Assistive devices can use local or cloud-based algorithmic detection and alerting of either or both of the subject person and a qualified healthcare professional (QHP) to review the glycemia data arid consider whether additional diagnostic testing, such as antibody testing or oral glucose tolerance testing based on the person’s trend over time from a state of euglycemia toward a state of dysglycemia detected in patterns of continuous glucose monitoring data or a model of such.
  • QHP healthcare professional
  • Augmentative devices can use local or cloud-based algorithmic quantification as an element of diagnosis and alerting either or both of the subject person and a QHP of an Aby+ person’s transition over time from a state of euglycemia to a state of dysglycemia based on paterns in continuous glucose monitoring data or a model of such.
  • Autonomous devices can use local or cloud based algorithmic diagnosis and recommendation to a QHP for preventative therapeutic intervention by taking into account the antibody status [Aby+ or Aby-] of the subject person and their transition over time from a state of euglycemia to a state of dysglycemia based on patterns in continuous glucose monitoring data or a model of such.
  • One type of device is based on a paper-based colorimetric lateral flow assay for measuring glucose from a drop of blood. Some such assays can be read by electronic devices, and some such electronic devices are connected to the internet and thereby able to upload results to a cloud server with personal data or other electronic health records.
  • FIG. 4 shows a block diagram of a system for monitoring people using devices connected through a network 40.
  • the network might be a local area network, such as one within a hospital, or the internet.
  • a first monitoring device 41 monitors a first person. It is controlled by and provides data to a mobile device 42. Smartphones, tablet computers, home digital assistant devices, and healthcare provider operated devices are some examples.
  • a second monitoring device 43 monitors a person. It could be the same or a different person.
  • the second monitoring device 43 includes built-in information collection and storage capabilities and one or more of display and alert capabilities. Smartwatches, health rings, and clip-on monitors are examples.
  • a third monitoring device 44 also monitors a person. It has no local user interface. It simply takes measurements and uploads them through the network 40.
  • the three monitoring devices communicate through the network 40 to a server 45.
  • the server includes one or more processors such as a central processing unit and graphics processing unit.
  • the server also includes storage such as a dynamic random access memory (DRAM) io store program code for the processors and data used for processing.
  • DRAM dynamic random access memory
  • the server 45 reads algorithm software instructions from a non-volatile storage device such as a magnetic hard disk drive or one or more Flash random access memory (RAM) chips.
  • the server 45 executes the algorithm software, which accesses data from a hard drive or Flash RAM storage device 47.
  • the database 47 and algorithm software 46 are stored in either fee same or different data storage devices.
  • the server 45 receives measurements and other data from the monitoring devices. It processes the data to identify monitoring devices that have measured a state of dysglycemia. hi some implementations, the state of glycemia that characterize dysglycemia are person-specific and therefore implemented as conditional on personal data stored in the database 47.
  • the server 45 sends to a clinical decision support platform 48 one or more of alerts, measurements, patterns in measurements, patterns, and other information learned from measurements by monitoring devices.
  • Some such platforms provide dashboards for periodic review and real-time alerts to physicians.
  • the server detects, for example, a person in dysglycemia, it can send an alert to a physician through the clinical decision support platform 48.
  • Some examples of such a device are a computer in a doctor’s office, a SMS pager, a smartphone or tablet computer app, or an email server.
  • the physician alert device can inform a physician, in various implementations, of the identity of a patient in dysglycemia, the fact that they are in such a state, a chart of their recent glycemia prior to entering dysglycemia, and information for contacting the patient
  • Physician alert devices in various settings, can be operated by physicians, nurses, hospital information technology (IT) services, or EHR providers.
  • the server 45 upon recognizing that a person is in dysglycemia, can send the same or similar information as provided to a physician alert system 48 to a mobile device used by a person or their family member. For example, monitor device 44 only provides data to the server 45, not to the person. That monitoring device 44 might be used by a child.
  • Another type of monitor is a wrist watch that estimates glycemia without puncturing the skin (non-invasive monitoring). That can be performed in multiple ways, one of which is to emit light into the skin and measure the intensity of light reflected back.
  • Some such smartwatches connect to smartphones or other devices that are connected to the internet and can upload glucose monitoring data to a cloud server.
  • Some smartwatches connect directly to the internet through a wireless communication network such as 5G or WiFi,
  • monitor is one implanted under the skin. It can measure glycemia in interstitial fluid and provide the measurement data to an electronic reader device wirelessly using a protocol such as near-field communication (NFC). Some such readers are connected to the internet and can upload data to the cloud.
  • NFC near-field communication
  • glucose monitoring assays, devices, and other ways of measuring glycemia can provide personal health data if sampled with a frequency higher than the frequency of transitions between euglycemia to dysglycemia in blood glycemia. At least 2x higher frequency is ideal due to the Shannon theorem, but higher frequencies are generally better. Indeed, to detect a trend in the pattern of glycemia, whether directly from patterns in the data or patterns from a model learned from the data, would require a much richer database, ie, with many more frequent measurements. Despite the sampling frequency and type of device, such approaches are known as continuous glucose monitoring (CGM).
  • CGM continuous glucose monitoring
  • the measurements needed to delect changes in glycemia can be captured using a CGM device. They could also be captured by other types of devices that provide a stream of frequent measurements with high enough sampling frequency to recognize variations that can predict unexpected changes in glycemia . For example, it would be necessary to detect an abnormally rapid rate of rise in blood glucose after a carbohydrate load, due to die slower response of the beta cells to secrete insulin.
  • Pattern analysis Whether connected to the cloud or processed locally, glucose monitoring assays, devices, and other ways of measuring glycemia can provide valuable personal health data if sampled with a frequency higher than the frequency of transitions between euglycemia to dysglycemia in blood glycemia. At least 2x higher frequency is ideal due to the Shannon theorem, but higher frequencies are generally better. Sampling once every 15 minutes is appropriate for some applications. Once every 5 minutes is appropriate for some applications. Other more or less frequent sampling periods are also appropriate for some applications.
  • FIG. 5 is a diagram of an example of a method for reporting a trend toward or onset of pathophysiologic dysglycemia. It begins with obtaining glucose monitoring data for a person.
  • glucose monitoring data may be obtained from an EHR system, a device vendor’s cloud server, zippie Health or Google Fit, directly from the computer memory within a device, or other sources.
  • a machine learning process 51 is applied to the data. This generates parameters for a model 52.
  • the model 52 takes in a specified time of day and, from that, can infer a pattern of glycemia.
  • die parameters are weights for a neural network, In some implementations, they are inputs to a formula.
  • the model 52 also takes, as input, one or more risk factors specific to the person. Examples of risk factors are the person’s family history of diabetes, whether they have tested positive for antibodies, and how many, and when, and whether they have genetic mutations frequently associated with diabetes.
  • the model 52 can take in glucose monitoring data from a different (generally later) time period and, in some implementations, additional data, to predict a pattern of glycemia.
  • the pattern can be daily fluctuations, peaks immediately after meals, or aggregate fluctuations.
  • the prediction is a process of inference by the machine learned model 52.
  • the model 52 outputs a predicted pattern of glycemia over a range of time. The method compares the prediction to actual glucose monitoring data received subsequently to training the model 52. The comparison is used to compute a deviation 53 of the actual from the predicted pattern.
  • Computing the deviation can include either or both of (a) computing a difference in amplitude between the predicted and measured glycemia data and (b) computing a difference in rate of change between the predicted and measured glycemia data.
  • the deviation may be reported directly to a QHP or their patient or patient 5 s caretaker. Alternatively or additionally, the deviations can be translated to an estimated probability of certain outcomes.
  • the resulting report can indicate a probability of the person testing positive in an OGTT test, the probability of a clinical diagnosis of pathophysiologic dysglycemia, the probability of a diagnosis of type 1 diabetes in stage 2, a probability of an event of ketoacidosis, and probabilities of other events or diagnoses.
  • FIG. 6 is a diagram of another example of a method for reporting pathophysiologic dysglycemia. It uses a pre-trained model 61 of patterns of glycemia, the model being conditioned on risk factors such as the ones described above.
  • the pre-trained model 61 is used with personspecific glucose monitoring data for a process of fine tuning 62 to determine model parameters. These are used to create a model 63 of patterns of glycemia for the person.
  • the model can take, as input, measured patterns of glycemia and, in some implementations, corresponding additional data.
  • the model 63 outputs predicted patterns of glycemia.
  • the computation of the prediction is a method of inference using fee machine learned model 63.
  • the predicted patterns can be compared to patterns in actual glucose monitoring data in a further step 64 to produce a report.
  • the report in various embodiments, can include the types of probabilistic predictions described above.
  • FIG. 7 is a table of examples of risk factors and the data types of values for each risk factor shown.
  • antibody test results and family history can either be positive, negative, or unknown.
  • the input can be a number representing a level in units of, for example, milligrams per deciliter (mg/dL) or unknown.
  • mg/dL milligrams per deciliter
  • a number of mg/dL can be represented as a model input, with, for example, an integer or floating point number.
  • genetic biomarkers can have three states of being normal (no mutation from the common human genome), mutated, or unknown. Genetic mutations can be detected by gene sequencing of the person being tested. In general, if any base pair of a necessary gene has a mutation, the protein for which it codes will be completely ineffective, and the protein that is normally coded at the location of that gene in the human genome will not be produced.
  • Another example risk factor is age, which can be represented with an integer number. Another example is gender as determined by the presence or absence of a human Y chromosome, which can be expressed as a single binary bit.
  • Another example risk factor is zip code, which is a sequence of characters (5 numbers in the US, 6 numbers and letters in many other countries). The zip code can be used as a key into a demographic lookup table of probabilities of various diagnoses and outcomes.
  • FIG. 8 is a diagram of another example method for reporting pathophysiologic dysglycemia. It receives historical glucose monitoring data for a specific person and uses the data as input to a training process 81. The training produces parameters for a model 82 that can predict patterns for real- time glycemia. The model 82 can be applied in real -time to glucose monitoring data from the person. The model can also, in some implementations, use the person’s risk factors and additional real-time data for the person.
  • the method of FIG. 8 received real-time glucose monitoring data.
  • the model 82 predicts patterns of glycemia levels. That can be compared, in real-time to the real-time monitoring data to compute a deviation 83 of tiie actual real-time glycemia levels from die predicted glycemia levels.
  • the level of de viation can be reported directly to the person or their QHP. Alternatively or additionally, the deviation can be translated to derived predictions or alerts.
  • the deviations can be used to produce warnings or alerts. Whereas some conventional systems give alerts when a user approaches a condition of dysglycemia, that can have unnecessary false alarms. False alarms cause users to develop a habit of ignoring alarms.
  • the implementation of FIG. 8 provides for more accurate compliance of patents and early warning of actual, serious, recent onset of or clinically meaningful progression in trends toward conditions of dysglycemia and related complications.
  • An additional feature of the example implementation of FIG. 8 is that the process of learning a person-specific model of patterns of glycemia can also produce parameters and models that can be used to configure devices such as CGM diagnostic devices and insulin pump therapeutic devices.
  • FIG. 9 is a diagram of another example of a method for reporting pathophysiologic dysglycemia. It includes a step 91 of training a machine -learned model 92 to predict patterns of glycemia from additional data captured during the same time period as the glucose monitoring training data. The method includes a similar step 93 of training a machine-learned model 94 to predict patterns of glycemia from person- specific glucose monitoring data and additional data from a second time period that is different from the first time period.
  • the accuracy of the first trained model 92 and second trained model 94 are higher if trained with more data.
  • To train with enough data for a clinically dependable prediction accuracy would require, in some implementations, data collected for a duration of at least 1 month or 30 days. Assuming 3 meals per day, that gives about 90 examples of peaks and decays of glycemia and, also, will span the occurrence of many other events which are factors in glycemia, such as strenuous exercise which might occur irregularly or an occasional span of a few meals all of which axe low in carbohydrates. More data is generally preferable.
  • the second time period of data used to train the second model begins at least 3 months or 90 days after the first time period of data used to train the first model, that is sufficient for some implementations to have clinically meaningful accuracy. Longer time periods for both CGM data and spacing between model training periods improves accuracy.
  • additional data can be used for various implementations. Some examples are time of meals (logged or captured by sensors such as glasses with cameras an object recognition image processing algorithm), contents of meals (logged or captured similarly), bodymovement such as number or frequency of steps, heart rate, blood oxygen concentration, blood pressure, body temperature, ambient temperature, sleep patterns, and neural activity, or the results of other diagnostic testing, such as Aby tests and the duration since they became positive.
  • such types of additional data can be captured by body- worn devices, captured by adjacent devices, logged manually by users, or accessed from other databases, e.g. via API from EHRs.
  • the capture of the additional data is especially useful if captured concurrently with the glucose monitoring data.
  • the concurrent capture of additional data can be instantaneously sampled with glycemia data or captured one or more times during a period of capturing the related glycemia data.
  • Well -trained models can predict patterns of glycemia with high accuracy for any given pattern of additional data.
  • Industry standard simulation patterns of the additional data may be agreed to give most accurate diagnoses.
  • Some simulation patterns might test large deviations in dietary sugar intake, large deviations in exercise regimens, or long-term averages.
  • the method runs the first model 92 with the simulation pattern in steps 95 and runs the second model 96 on the simulation pattern in step 96 to produce simulated patterns of glycemia.
  • a further step 97 computes a deviation in the simulated patterns. The deviations can be used for a step of creating a report 98 as described with respect to the examples of FIG. 5 and FIG. 6.
  • the report creation 98 can also be conditioned on risk factor parameters.
  • a model of risk factor parameters 99 can be trained from data across a large population. Applying the risk factor model 99 to the person -specific risk factor produces parameters that can be used to condition die report creation 98. Accordingly, an accurate person-specific report is produced that a QHP, patient, or their caretaker can use to develop a care plan.
  • Some implementations include models of large populations in conjunction with person- specific models.
  • the risk factor model 99 of FIG. 9 is one example of a population-wide model, and the pre-trained model 61 of FIG. 6 is another.
  • Such models can, in various implementations, be trained using a corpus of real world evidence (RWE) associated with medical diagnoses.
  • RWE real world evidence
  • Such corpora are available within some private health systems such as Kaiser Pennanente or national health systems such as tire British National Health Service (NIIS).
  • the corpus of medical data is labeled with attributes of the subject person including one or more of age, antibody test status (if known), time since the Aby test became positive, OGTT test results (if known), genetic risk factors (if known), and other risk factors.
  • attributes of the subject person including one or more of age, antibody test status (if known), time since the Aby test became positive, OGTT test results (if known), genetic risk factors (if known), and other risk factors.
  • a well labeled corpus enables a model to be trained that is easily made conditional based on labeled attributes of the data. Results can be filtered by conditional labels. But models of statistical regression can also be used with non-binary label values for attributes such as average heart rate and estimated sugar intake with meals.
  • HLA Human leukocyte antigen
  • MHC major histocompatibility complex
  • HLAs encode cell surface proteins that regulate the immune system. Mutations in the HLA coding segments of DNA are known to cause type 1 diabetes and some other autoimmune diseases. Mutations in HLA-DR3 and HLA- DR4 are relatively common in people with European descendants. Mutations in HLA - DR7 are relatively common in people with African descendants. Mutations in HLA-DR9 are relatively common in people with Japanese descendants. Personal genome sequencing can detect, with high certainty, whether a person has such mutations or not, which is a binary risk factor. HLA mutations can be detected from gene sequencing around birth, any time afterwards, and even in utero. [0084] The presence, absence, or unknown state of HLA mutations can be input factors to a statistical model trained to predict likelihood and severity of type 1 diabetes. Such models can be trained from corpora of records of type 1 diabetes symptoms and gene mutations across large populations.
  • Other yet unidentified genetic mutations may also cause or create a predisposition to type 1 diabetes. Such mutations may also be used in such a model.
  • Some other risk factors can be measured as non-binary levels. For example, behavioral and dynamic risk factors such as measured or estimated amounts of different types of sugars and other carbohydrates eaten, percent body fat, co-morbidities like polycystic ovarian syndrome, amount of exercise as identifiable through heart rate or step count, and length of time spent logged in to devices having their screens turned on.
  • behavioral and dynamic risk factors such as measured or estimated amounts of different types of sugars and other carbohydrates eaten, percent body fat, co-morbidities like polycystic ovarian syndrome, amount of exercise as identifiable through heart rate or step count, and length of time spent logged in to devices having their screens turned on.
  • risk factors useful for training conditional models include data from other body-worn or implanted sensors such as meal type recognition by logging or camera image processing, timing of meals, blood oxygen concentration, blood pressure, body temperature, ambient temperature, sleep patterns, and neural activity.
  • FIG. 10 shows a high level flowchart of one possible care pathway.
  • a doctor such as a pediatrician, screens a patient with an antibody diagnostic (Dx) test. If the patient is Aby+, the doctor may recommend or prescribe CGM for the patient.
  • Dx antibody diagnostic
  • the CGM device uploads test information to a cloud based server.
  • the server uses a multiplicity of CGM glycemia samples, and potentially additional information, to train a patient-specific model of patterns of glycemia. In particular, some trained models predict the patient’s glycemic response to meals.
  • the cloud server can also use real-time CGM readings of glycemia to recognize dysglycemia and alert a physician such as a pediatrician or other QHP.
  • the QHP in response to a model predicting pathophysiologic dysglycemia, can prescribe a therapy such as a prescription to the pharmaceutical therapeutic Teplizumab.
  • the prescription step can be done, even if few or no incidents of acute dysglycemia have occurred. This can prevent DKA, long before it would occur under other standards of care.
  • the patient care pathway might include, upon a cloud based prediction of pathophysiologic dysglycemia, a step 104 of a physician, such as a pediatrician, a primary care physician, or general practitioner evaluating and monitoring the patient.
  • the evaluation may include an OGTT, Depending on the results of the evaluation and monitoring, the physician may refer the patient to a specialist such as an endocrinologist.
  • a specialist such as an endocrinologist.
  • the choice of endocrinologist may involve selecting from one or more endocrinologists participating within a physician network.
  • an additional step 105 the endocrinologist performs additional evaluations and monitoring. This may include an OGTT, genetic testing, or other screening. The endocrinologist may then prescribe a therapy step 103 such as a prescription for Teplizumab.
  • FIG. 11 shows an example of a standard of care pathway using detection of deviation, interpretation, and a conclusion of diagnosis and recommended intervention.
  • a doctor such as a pediatrician, screens a patient with an antibody diagnostic (Dx) test. If the patient is Aby+, the doctor may recommend or prescribe CGM for the patient.
  • Dx antibody diagnostic
  • the CGM device uploads test information to a cloud based server.
  • the server uses a multiplicity of CGM glycemia samples, and potentially additional information, to train a patient -specific model of patterns of glycemia.
  • some trained models predict die patient’s glycemic response to meals
  • lite cloud server can also use real-time CGM readings of glycemia to recognize dysglycemia and alert a physician such as a pediatrician or other QHP.
  • an algorithm on the cloud server compares subsequent CGM data to a pattern predicted by the trained model to detect the presence or measure the amount of deviation between the measured data and the predicted pattern.
  • an algorithm interprets the clinical meaningfulness of the deviation.
  • an algorithm draws a conclusion from the interpretation. The conclusion could be, for example, that there is no disease detected or that the person has type 1 diabetes at stage 2.
  • an algorithm recommends an intervention. Hie recommended intervention can be selected from a set of rules based on the measured data.
  • the rale might include non-binary information, such as a dosage level for a preventative pharmaceutical treatment.
  • care pathway showing detection of the transition over time from a state of euglycemia to a state of dysglycemia based on patterns from a learned model may be used for the care pathways described above.
  • Another standard of care computes a tripartite likelihood/confidence score from the model, such that, for patients with lower scores, physicians simply continue on the AICGM. For middle scores, patients are sent for OGTT, and patients with tipper scores get therapeutic prescriptions directly.
  • pathophysiologic dysglycemia can be used asynchronously or synchronously with evaluation and management. Recognizing pathophysiologic trends recognizes progression toward stages of diabetes which approach a threshold at which the pathophysiologic state might warrant preventive intervention
  • An additional benefit of the new standards of care enabled by early recognition of pathophysiologic dysglycemia is to delay insulin deficiency associated with stages of diabetes. This is an improvement over therapeutic intervention by replacing insulin or administering metformin to enhance insulin responsiveness.
  • the computer processors may be central processing units (CPU), graphics processing units (GPU), digital signal processors (DSP), or others.
  • CPU central processing units
  • GPU graphics processing units
  • DSP digital signal processors
  • Such processors are typically implemented in chips such as system-on-chip (SoC) devices or field programmable gale arrays (FPGA).
  • SoC system-on-chip
  • FPGA field programmable gale arrays

Landscapes

  • Health & Medical Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Public Health (AREA)
  • Medical Informatics (AREA)
  • Biomedical Technology (AREA)
  • General Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Epidemiology (AREA)
  • Primary Health Care (AREA)
  • Pathology (AREA)
  • Physics & Mathematics (AREA)
  • Databases & Information Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • Surgery (AREA)
  • Animal Behavior & Ethology (AREA)
  • Veterinary Medicine (AREA)
  • Psychiatry (AREA)
  • Artificial Intelligence (AREA)
  • Signal Processing (AREA)
  • Optics & Photonics (AREA)
  • Physiology (AREA)
  • Emergency Medicine (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Business, Economics & Management (AREA)
  • Business, Economics & Management (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Medicinal Chemistry (AREA)
  • Chemical & Material Sciences (AREA)
  • Investigating Or Analysing Biological Materials (AREA)
  • Measuring And Recording Apparatus For Diagnosis (AREA)

Abstract

Des conditions de dysglycémie pathophysiologique, associées au diabète de type 1, peuvent être détectées précocement, y compris chez les enfants. Des données de glycémie sont collectées par des dispositifs de surveillance continue du glucose connectés à un serveur cloud. Le serveur apprend des motifs spécifiques à une personne de glycémie et entraîne un modèle en conséquence, qui peut servir de ligne de base pour une comparaison ou en tant que base pour la prédiction de profils futurs. De tels modèles sont conditionnés par des facteurs de risque et diverses activités simultanées telles que l'exercice et la consommation de sucres. Des modèles spécifiques à une personne entraînés à différentes périodes de temps donnent différents motifs simulés de glycémie. Des écarts entre les motifs simulés, ou entre les motifs prédits et réels, peuvent indiquer la progression de la dysglycémie pathophysiologique et du diabète de type 1. Ces modèles et comparaisons peuvent en outre être interprétés par le logiciel pour obtenir une recommandation rapportée pour un test de diagnostic supplémentaire ou une conclusion rapportée de diagnostic et de recommandation d'intervention préventive.
PCT/US2024/047305 2023-09-18 2024-09-18 Reconnaissance précoce de changement d'état pathophysiologique de dysglycémie Pending WO2025064557A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
US18/469,360 US20250090103A1 (en) 2023-09-18 2023-09-18 Early recognition of change to pathophysiologic state of dysglycemia
US18/469,360 2023-09-18

Publications (1)

Publication Number Publication Date
WO2025064557A1 true WO2025064557A1 (fr) 2025-03-27

Family

ID=94977265

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/US2024/047305 Pending WO2025064557A1 (fr) 2023-09-18 2024-09-18 Reconnaissance précoce de changement d'état pathophysiologique de dysglycémie

Country Status (2)

Country Link
US (1) US20250090103A1 (fr)
WO (1) WO2025064557A1 (fr)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN120473075B (zh) * 2025-07-17 2025-09-30 中国人民解放军北部战区总医院 一种临床护理中康复方案动态推荐方法及系统

Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7118919B2 (en) * 1998-05-06 2006-10-10 Isotechnika Inc. 13C glucose breath test for the diagnosis of diabetic indications and monitoring glycemic control
US20200118647A1 (en) * 2018-10-12 2020-04-16 Ancestry.Com Dna, Llc Phenotype trait prediction with threshold polygenic risk score
WO2021040878A1 (fr) * 2019-08-30 2021-03-04 Massachusetts Institute Of Technology Surveillance non invasive du glucose par spectroscopie raman
US20210077719A1 (en) * 2019-09-13 2021-03-18 Insulet Corporation Blood glucose rate of change modulation of meal and correction insulin bolus quantity
US20210187197A1 (en) * 2019-12-20 2021-06-24 Insulet Corporation Techniques for improved automatic drug delivery performance using delivery tendencies from past delivery history and use patterns
US20210338116A1 (en) * 2020-04-29 2021-11-04 Dexcom, Inc. Hypoglycemic event prediction using machine learning
US20220142521A1 (en) * 2020-11-10 2022-05-12 Ascensia Diabetes Care Holdings Ag Methods and apparatus for predicting whether and when a hypo/hyper analyte concentration event will occur
WO2023034542A1 (fr) * 2021-09-01 2023-03-09 Integrated Medical Sensors, Inc. Circuit de capteur intégré monolithique implantable filaire
US11615392B2 (en) * 2020-05-01 2023-03-28 Capital One Services, Llc Systems and methods for using information from wearable devices
KR20230060312A (ko) * 2021-10-27 2023-05-04 (주)스톤랩 시각 인공지능 기반 항체 신속 테스트키트 진단결과 식별 및 인증 시스템 및 그 방법

Family Cites Families (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP2494364A1 (fr) * 2009-10-29 2012-09-05 Tethys Bioscience, Inc. Biomarqueurs de protéines et de lipides améliorant considérablement la prédiction du diabète de type 2
US8868185B2 (en) * 2011-10-31 2014-10-21 Medtronic, Inc. Use of thoracic and extra-thoracic impedance for diagnostic monitoring
CN110753967A (zh) * 2017-06-15 2020-02-04 诺和诺德股份有限公司 基于患者概况的胰岛素滴定算法
US11929171B2 (en) * 2018-10-18 2024-03-12 The Board Of Trustees Of The Leland Stanford Junior University Methods for evaluation and treatment of glycemic dysregulation and atherosclerotic cardiovascular disease and applications thereof
WO2021030637A1 (fr) * 2019-08-13 2021-02-18 Twin Health, Inc. Amélioration de la santé métabolique à l'aide d'une plateforme de traitement de précision activée par une technologie jumelée numérique du corps entier
KR102044419B1 (ko) * 2019-08-16 2019-11-14 주식회사 클리노믹스 유전자 검사를 통한 사용자 맞춤형 음료추천 시스템 및 그 구동방법
US11696706B2 (en) * 2019-12-16 2023-07-11 Dexcom, Inc. Multi-state engagement with continuous glucose monitoring systems
US12496398B2 (en) * 2021-05-28 2025-12-16 Insulet Corporation Threshold based automatic glucose control response
AU2022379460A1 (en) * 2021-10-27 2024-05-23 Dexcom, Inc. Disease prediction using analyte measurement features and machine learning

Patent Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7118919B2 (en) * 1998-05-06 2006-10-10 Isotechnika Inc. 13C glucose breath test for the diagnosis of diabetic indications and monitoring glycemic control
US20200118647A1 (en) * 2018-10-12 2020-04-16 Ancestry.Com Dna, Llc Phenotype trait prediction with threshold polygenic risk score
WO2021040878A1 (fr) * 2019-08-30 2021-03-04 Massachusetts Institute Of Technology Surveillance non invasive du glucose par spectroscopie raman
US20210077719A1 (en) * 2019-09-13 2021-03-18 Insulet Corporation Blood glucose rate of change modulation of meal and correction insulin bolus quantity
US20210187197A1 (en) * 2019-12-20 2021-06-24 Insulet Corporation Techniques for improved automatic drug delivery performance using delivery tendencies from past delivery history and use patterns
US20210338116A1 (en) * 2020-04-29 2021-11-04 Dexcom, Inc. Hypoglycemic event prediction using machine learning
US11615392B2 (en) * 2020-05-01 2023-03-28 Capital One Services, Llc Systems and methods for using information from wearable devices
US20220142521A1 (en) * 2020-11-10 2022-05-12 Ascensia Diabetes Care Holdings Ag Methods and apparatus for predicting whether and when a hypo/hyper analyte concentration event will occur
WO2023034542A1 (fr) * 2021-09-01 2023-03-09 Integrated Medical Sensors, Inc. Circuit de capteur intégré monolithique implantable filaire
KR20230060312A (ko) * 2021-10-27 2023-05-04 (주)스톤랩 시각 인공지능 기반 항체 신속 테스트키트 진단결과 식별 및 인증 시스템 및 그 방법

Also Published As

Publication number Publication date
US20250090103A1 (en) 2025-03-20

Similar Documents

Publication Publication Date Title
AU2024202376B2 (en) Database management and graphical user interfaces for measurements collected by analyzing blood
Afsaneh et al. Recent applications of machine learning and deep learning models in the prediction, diagnosis, and management of diabetes: a comprehensive review
KR102400740B1 (ko) 사용자의 건강상태 모니터링 시스템 및 이의 분석 방법
CN112102953B (zh) 个性化糖尿病健康管理系统、设备及存储介质
CN114207737A (zh) 用于生物监测和血糖预测的系统及相关联方法
US20110201901A1 (en) Systems and Methods for Predicting Patient Health Problems and Providing Timely Intervention
Saheed et al. Modified bi-directional long short-term memory and hyperparameter tuning of supervised machine learning models for cardiovascular heart disease prediction in mobile cloud environment
JP2024513618A (ja) 感染症及び敗血症の個別化された予測のための方法及びシステム
CN104207753A (zh) 个人健康监视系统
CN105279362A (zh) 个人健康监视系统
US20240225525A1 (en) Thyroid function monitoring method according to medication, and monitoring server and user terminal performing the same
CN113679348A (zh) 血糖预测方法、血糖预测装置、设备及存储介质
US20240274294A1 (en) A method and system for risk prediction of neonatal infection
WO2025064557A1 (fr) Reconnaissance précoce de changement d'état pathophysiologique de dysglycémie
US12315604B2 (en) Recurring remote monitoring with real-time exchange to analyze health data and generate action plans
Roy et al. OBESEYE: Interpretable Diet Recommender for Obesity Management using Machine Learning and Explainable AI
US20250218598A1 (en) Systems and methods for continuous glucose monitoring outcome predictions
WO2025224714A1 (fr) Système et procédé de génération d'instructions de soins de patient à l'aide de données de capteur en temps réel
Bosoni Temporal data analytics for diabetes monitoring with applications on pediatric and adult patients
Gunay Modelling of sleep behaviors of patients with mood disorders
WO2025006393A1 (fr) Alertes prédictives d'hyperglycémie longue
KR20250002870A (ko) 주치의 의료 분석 구독 서비스를 수행하는 방법 및 컴퓨팅 장치
Chatrati et al. Computer and Information Sciences

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 24869092

Country of ref document: EP

Kind code of ref document: A1